A direct thrust control method of scramjet based on deep reinforcement learning

In view of the complex plant of scramjet with strong nonlinear characteristics， a deep reinforcement learning control method for scramjet direct thrust control is proposed， considering the requirement of fast response and safety boundary limitation during control. Based on the one-dimensional model of the scramjet， the thrust characteristics are analyzed， and the thrust estimator of the scramjet is established via BP neural network. Based on the deep reinforcement learning control method， the thrust control strategy of scramjet is proposed and its parameters are optimized. By setting the penalty function， the constrained deep reinforcement learning control algorithm is designed to realize the limited parameter protection control of the scramjet. The simulation results show that the proposed method has better dynamic and steady-state performances than the traditional PID control method.